3 * This file is part of the GROMACS molecular simulation package.
5 * Copyright (c) 2012,2013, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
36 #error This file must be processed with the Gromacs pre-preprocessor
38 /* #if INCLUDE_HEADER */
45 #include "../nb_kernel.h"
46 #include "types/simple.h"
50 #include "gmx_math_x86_sse4_1_double.h"
51 #include "kernelutil_x86_sse4_1_double.h"
54 /* ## List of variables set by the generating script: */
56 /* ## Setttings that apply to the entire kernel: */
57 /* ## KERNEL_ELEC: String, choice for electrostatic interactions */
58 /* ## KERNEL_VDW: String, choice for van der Waals interactions */
59 /* ## KERNEL_NAME: String, name of this kernel */
60 /* ## KERNEL_VF: String telling if we calculate potential, force, or both */
61 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
63 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops: */
64 /* ## PARTICLES_I[]/ Arrays with lists of i/j particles to use in kernel. It is */
65 /* ## PARTICLES_J[]: just [0] for particle geometry, but can be longer for water */
66 /* ## PARTICLES_ELEC_I[]/ Arrays with lists of i/j particle that have electrostatics */
67 /* ## PARTICLES_ELEC_J[]: interactions that should be calculated in this kernel. */
68 /* ## PARTICLES_VDW_I[]/ Arrays with the list of i/j particle that have VdW */
69 /* ## PARTICLES_VDW_J[]: interactions that should be calculated in this kernel. */
71 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle) */
72 /* ## PAIRS_IJ[]: Array with (i,j) tuples of pairs for which interactions */
73 /* ## should be calculated in this kernel. Zero-charge particles */
74 /* ## do not have interactions with particles without vdw, and */
75 /* ## Vdw-only interactions are not evaluated in a no-vdw-kernel. */
76 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
77 /* ## For each i-j pair, the element [I][J] is a list of strings */
78 /* ## defining properties/flags of this interaction. Examples */
79 /* ## include 'electrostatics'/'vdw' if that type of interaction */
80 /* ## should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values */
81 /* ## are needed, and 'exactcutoff' or 'shift','switch' to */
82 /* ## decide if the force/potential should be modified. This way */
83 /* ## we only calculate values absolutely needed for each case. */
85 /* ## Calculate the size and offset for (merged/interleaved) table data */
88 * Gromacs nonbonded kernel: {KERNEL_NAME}
89 * Electrostatics interaction: {KERNEL_ELEC}
90 * VdW interaction: {KERNEL_VDW}
91 * Geometry: {GEOMETRY_I}-{GEOMETRY_J}
92 * Calculate force/pot: {KERNEL_VF}
96 (t_nblist * gmx_restrict nlist,
97 rvec * gmx_restrict xx,
98 rvec * gmx_restrict ff,
99 t_forcerec * gmx_restrict fr,
100 t_mdatoms * gmx_restrict mdatoms,
101 nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
102 t_nrnb * gmx_restrict nrnb)
104 /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
105 /* ## so there is no point in going to extremes to exclude variables that are not needed. */
106 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
107 * just 0 for non-waters.
108 * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
109 * jnr indices corresponding to data put in the four positions in the SIMD register.
111 int i_shift_offset,i_coord_offset,outeriter,inneriter;
112 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
114 int j_coord_offsetA,j_coord_offsetB;
115 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
117 real *shiftvec,*fshift,*x,*f;
118 __m128d tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
119 /* #for I in PARTICLES_I */
121 __m128d ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
123 /* #for J in PARTICLES_J */
124 int vdwjidx{J}A,vdwjidx{J}B;
125 __m128d jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
127 /* #for I,J in PAIRS_IJ */
128 __m128d dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
130 /* #if KERNEL_ELEC != 'None' */
131 __m128d velec,felec,velecsum,facel,crf,krf,krf2;
134 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
136 __m128d vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
137 __m128d minushalf = _mm_set1_pd(-0.5);
138 real *invsqrta,*dvda,*gbtab;
140 /* #if KERNEL_VDW != 'None' */
142 __m128d rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
145 __m128d one_sixth = _mm_set1_pd(1.0/6.0);
146 __m128d one_twelfth = _mm_set1_pd(1.0/12.0);
148 /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
150 __m128i ifour = _mm_set1_epi32(4);
151 __m128d rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
154 /* #if 'Ewald' in KERNEL_ELEC */
156 __m128d ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
159 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
160 __m128d rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
161 real rswitch_scalar,d_scalar;
163 __m128d dummy_mask,cutoff_mask;
164 __m128d signbit = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
165 __m128d one = _mm_set1_pd(1.0);
166 __m128d two = _mm_set1_pd(2.0);
172 jindex = nlist->jindex;
174 shiftidx = nlist->shift;
176 shiftvec = fr->shift_vec[0];
177 fshift = fr->fshift[0];
178 /* #if KERNEL_ELEC != 'None' */
179 facel = _mm_set1_pd(fr->epsfac);
180 charge = mdatoms->chargeA;
181 /* #if 'ReactionField' in KERNEL_ELEC */
182 krf = _mm_set1_pd(fr->ic->k_rf);
183 krf2 = _mm_set1_pd(fr->ic->k_rf*2.0);
184 crf = _mm_set1_pd(fr->ic->c_rf);
187 /* #if KERNEL_VDW != 'None' */
188 nvdwtype = fr->ntype;
190 vdwtype = mdatoms->typeA;
193 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
194 vftab = kernel_data->table_elec_vdw->data;
195 vftabscale = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
196 /* #elif 'Table' in KERNEL_ELEC */
197 vftab = kernel_data->table_elec->data;
198 vftabscale = _mm_set1_pd(kernel_data->table_elec->scale);
199 /* #elif 'Table' in KERNEL_VDW */
200 vftab = kernel_data->table_vdw->data;
201 vftabscale = _mm_set1_pd(kernel_data->table_vdw->scale);
204 /* #if 'Ewald' in KERNEL_ELEC */
205 sh_ewald = _mm_set1_pd(fr->ic->sh_ewald);
206 /* #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
207 ewtab = fr->ic->tabq_coul_F;
208 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
209 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
211 ewtab = fr->ic->tabq_coul_FDV0;
212 ewtabscale = _mm_set1_pd(fr->ic->tabq_scale);
213 ewtabhalfspace = _mm_set1_pd(0.5/fr->ic->tabq_scale);
217 /* #if KERNEL_ELEC=='GeneralizedBorn' */
218 invsqrta = fr->invsqrta;
220 gbtabscale = _mm_set1_pd(fr->gbtab.scale);
221 gbtab = fr->gbtab.data;
222 gbinvepsdiff = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
225 /* #if 'Water' in GEOMETRY_I */
226 /* Setup water-specific parameters */
227 inr = nlist->iinr[0];
228 /* #for I in PARTICLES_ELEC_I */
229 iq{I} = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+{I}]));
231 /* #for I in PARTICLES_VDW_I */
232 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
236 /* #if 'Water' in GEOMETRY_J */
237 /* #for J in PARTICLES_ELEC_J */
238 jq{J} = _mm_set1_pd(charge[inr+{J}]);
240 /* #for J in PARTICLES_VDW_J */
241 vdwjidx{J}A = 2*vdwtype[inr+{J}];
243 /* #for I,J in PAIRS_IJ */
244 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
245 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
247 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
248 c6_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
249 c12_{I}{J} = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
254 /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
255 /* #if KERNEL_ELEC!='None' */
256 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
257 rcutoff_scalar = fr->rcoulomb;
259 rcutoff_scalar = fr->rvdw;
261 rcutoff = _mm_set1_pd(rcutoff_scalar);
262 rcutoff2 = _mm_mul_pd(rcutoff,rcutoff);
265 /* #if KERNEL_MOD_VDW=='PotentialShift' */
266 sh_vdw_invrcut6 = _mm_set1_pd(fr->ic->sh_invrc6);
267 rvdw = _mm_set1_pd(fr->rvdw);
270 /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
271 /* #if KERNEL_MOD_ELEC=='PotentialSwitch' */
272 rswitch_scalar = fr->rcoulomb_switch;
273 rswitch = _mm_set1_pd(rswitch_scalar);
275 rswitch_scalar = fr->rvdw_switch;
276 rswitch = _mm_set1_pd(rswitch_scalar);
278 /* Setup switch parameters */
279 d_scalar = rcutoff_scalar-rswitch_scalar;
280 d = _mm_set1_pd(d_scalar);
281 swV3 = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
282 swV4 = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
283 swV5 = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
284 /* #if 'Force' in KERNEL_VF */
285 swF2 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
286 swF3 = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
287 swF4 = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
291 /* Avoid stupid compiler warnings */
296 /* ## Keep track of the floating point operations we issue for reporting! */
297 /* #define OUTERFLOPS 0 */
301 /* Start outer loop over neighborlists */
302 for(iidx=0; iidx<nri; iidx++)
304 /* Load shift vector for this list */
305 i_shift_offset = DIM*shiftidx[iidx];
307 /* Load limits for loop over neighbors */
308 j_index_start = jindex[iidx];
309 j_index_end = jindex[iidx+1];
311 /* Get outer coordinate index */
313 i_coord_offset = DIM*inr;
315 /* Load i particle coords and add shift vector */
316 /* #if GEOMETRY_I == 'Particle' */
317 gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
318 /* #elif GEOMETRY_I == 'Water3' */
319 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
320 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
321 /* #elif GEOMETRY_I == 'Water4' */
322 /* #if 0 in PARTICLES_I */
323 gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
324 &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
326 gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
327 &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
331 /* #if 'Force' in KERNEL_VF */
332 /* #for I in PARTICLES_I */
333 fix{I} = _mm_setzero_pd();
334 fiy{I} = _mm_setzero_pd();
335 fiz{I} = _mm_setzero_pd();
339 /* ## For water we already preloaded parameters at the start of the kernel */
340 /* #if not 'Water' in GEOMETRY_I */
341 /* Load parameters for i particles */
342 /* #for I in PARTICLES_ELEC_I */
343 iq{I} = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+{I}));
344 /* #define OUTERFLOPS OUTERFLOPS+1 */
345 /* #if KERNEL_ELEC=='GeneralizedBorn' */
346 isai{I} = _mm_load1_pd(invsqrta+inr+{I});
349 /* #for I in PARTICLES_VDW_I */
350 vdwioffset{I} = 2*nvdwtype*vdwtype[inr+{I}];
354 /* #if 'Potential' in KERNEL_VF */
355 /* Reset potential sums */
356 /* #if KERNEL_ELEC != 'None' */
357 velecsum = _mm_setzero_pd();
359 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
360 vgbsum = _mm_setzero_pd();
362 /* #if KERNEL_VDW != 'None' */
363 vvdwsum = _mm_setzero_pd();
366 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
367 dvdasum = _mm_setzero_pd();
370 /* #for ROUND in ['Loop','Epilogue'] */
372 /* #if ROUND =='Loop' */
373 /* Start inner kernel loop */
374 for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
376 /* ## First round is normal loop (next statement resets indentation) */
383 /* ## Second round is epilogue */
385 /* #define INNERFLOPS 0 */
387 /* #if ROUND =='Loop' */
388 /* Get j neighbor index, and coordinate index */
391 j_coord_offsetA = DIM*jnrA;
392 j_coord_offsetB = DIM*jnrB;
394 /* load j atom coordinates */
395 /* #if GEOMETRY_J == 'Particle' */
396 gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
398 /* #elif GEOMETRY_J == 'Water3' */
399 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
400 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
401 /* #elif GEOMETRY_J == 'Water4' */
402 /* #if 0 in PARTICLES_J */
403 gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
404 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
405 &jy2,&jz2,&jx3,&jy3,&jz3);
407 gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
408 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
413 j_coord_offsetA = DIM*jnrA;
415 /* load j atom coordinates */
416 /* #if GEOMETRY_J == 'Particle' */
417 gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
419 /* #elif GEOMETRY_J == 'Water3' */
420 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
421 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
422 /* #elif GEOMETRY_J == 'Water4' */
423 /* #if 0 in PARTICLES_J */
424 gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
425 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
426 &jy2,&jz2,&jx3,&jy3,&jz3);
428 gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
429 &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
434 /* Calculate displacement vector */
435 /* #for I,J in PAIRS_IJ */
436 dx{I}{J} = _mm_sub_pd(ix{I},jx{J});
437 dy{I}{J} = _mm_sub_pd(iy{I},jy{J});
438 dz{I}{J} = _mm_sub_pd(iz{I},jz{J});
439 /* #define INNERFLOPS INNERFLOPS+3 */
442 /* Calculate squared distance and things based on it */
443 /* #for I,J in PAIRS_IJ */
444 rsq{I}{J} = gmx_mm_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
445 /* #define INNERFLOPS INNERFLOPS+5 */
448 /* #for I,J in PAIRS_IJ */
449 /* #if 'rinv' in INTERACTION_FLAGS[I][J] */
450 rinv{I}{J} = gmx_mm_invsqrt_pd(rsq{I}{J});
451 /* #define INNERFLOPS INNERFLOPS+5 */
455 /* #for I,J in PAIRS_IJ */
456 /* #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
457 /* # if 'rinv' not in INTERACTION_FLAGS[I][J] */
458 rinvsq{I}{J} = gmx_mm_inv_pd(rsq{I}{J});
459 /* #define INNERFLOPS INNERFLOPS+4 */
461 rinvsq{I}{J} = _mm_mul_pd(rinv{I}{J},rinv{I}{J});
462 /* #define INNERFLOPS INNERFLOPS+1 */
467 /* #if not 'Water' in GEOMETRY_J */
468 /* Load parameters for j particles */
469 /* #for J in PARTICLES_ELEC_J */
470 /* #if ROUND =='Loop' */
471 jq{J} = gmx_mm_load_2real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J});
473 jq{J} = _mm_load_sd(charge+jnrA+{J});
475 /* #if KERNEL_ELEC=='GeneralizedBorn' */
476 /* #if ROUND =='Loop' */
477 isaj{J} = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
479 isaj{J} = _mm_load_sd(invsqrta+jnrA+{J});
483 /* #for J in PARTICLES_VDW_J */
484 vdwjidx{J}A = 2*vdwtype[jnrA+{J}];
485 /* #if ROUND =='Loop' */
486 vdwjidx{J}B = 2*vdwtype[jnrB+{J}];
491 /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
492 /* #for J in PARTICLES_J */
493 fjx{J} = _mm_setzero_pd();
494 fjy{J} = _mm_setzero_pd();
495 fjz{J} = _mm_setzero_pd();
499 /* #for I,J in PAIRS_IJ */
501 /**************************
502 * CALCULATE INTERACTIONS *
503 **************************/
505 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
506 /* ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
507 if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
509 /* #if 0 ## this and the next two lines is a hack to maintain auto-indentation in template file */
512 /* #define INNERFLOPS INNERFLOPS+1 */
515 /* #if 'r' in INTERACTION_FLAGS[I][J] */
516 r{I}{J} = _mm_mul_pd(rsq{I}{J},rinv{I}{J});
517 /* #define INNERFLOPS INNERFLOPS+1 */
520 /* ## For water geometries we already loaded parameters at the start of the kernel */
521 /* #if not 'Water' in GEOMETRY_J */
522 /* Compute parameters for interactions between i and j atoms */
523 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
524 qq{I}{J} = _mm_mul_pd(iq{I},jq{J});
525 /* #define INNERFLOPS INNERFLOPS+1 */
527 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
528 /* #if ROUND == 'Loop' */
529 gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,
530 vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
532 gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,&c6_{I}{J},&c12_{I}{J});
537 /* #if 'table' in INTERACTION_FLAGS[I][J] */
538 /* Calculate table index by multiplying r with table scale and truncate to integer */
539 rt = _mm_mul_pd(r{I}{J},vftabscale);
540 vfitab = _mm_cvttpd_epi32(rt);
541 vfeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
542 /* #define INNERFLOPS INNERFLOPS+4 */
543 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
544 /* ## 3 tables, 4 data per point: multiply index by 12 */
545 vfitab = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
546 /* #elif 'Table' in KERNEL_ELEC */
547 /* ## 1 table, 4 data per point: multiply index by 4 */
548 vfitab = _mm_slli_epi32(vfitab,2);
549 /* #elif 'Table' in KERNEL_VDW */
550 /* ## 2 tables, 4 data per point: multiply index by 8 */
551 vfitab = _mm_slli_epi32(vfitab,3);
555 /* ## ELECTROSTATIC INTERACTIONS */
556 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
558 /* #if KERNEL_ELEC=='Coulomb' */
560 /* COULOMB ELECTROSTATICS */
561 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
562 /* #define INNERFLOPS INNERFLOPS+1 */
563 /* #if 'Force' in KERNEL_VF */
564 felec = _mm_mul_pd(velec,rinvsq{I}{J});
565 /* #define INNERFLOPS INNERFLOPS+2 */
568 /* #elif KERNEL_ELEC=='ReactionField' */
570 /* REACTION-FIELD ELECTROSTATICS */
571 /* #if 'Potential' in KERNEL_VF */
572 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_add_pd(rinv{I}{J},_mm_mul_pd(krf,rsq{I}{J})),crf));
573 /* #define INNERFLOPS INNERFLOPS+4 */
575 /* #if 'Force' in KERNEL_VF */
576 felec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
577 /* #define INNERFLOPS INNERFLOPS+3 */
580 /* #elif KERNEL_ELEC=='GeneralizedBorn' */
582 /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
583 isaprod = _mm_mul_pd(isai{I},isaj{J});
584 gbqqfactor = _mm_xor_pd(signbit,_mm_mul_pd(qq{I}{J},_mm_mul_pd(isaprod,gbinvepsdiff)));
585 gbscale = _mm_mul_pd(isaprod,gbtabscale);
586 /* #define INNERFLOPS INNERFLOPS+5 */
588 /* Calculate generalized born table index - this is a separate table from the normal one,
589 * but we use the same procedure by multiplying r with scale and truncating to integer.
591 rt = _mm_mul_pd(r{I}{J},gbscale);
592 gbitab = _mm_cvttpd_epi32(rt);
593 gbeps = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
594 gbitab = _mm_slli_epi32(gbitab,2);
596 Y = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
597 /* #if ROUND == 'Loop' */
598 F = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
600 F = _mm_setzero_pd();
602 GMX_MM_TRANSPOSE2_PD(Y,F);
603 G = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
604 /* #if ROUND == 'Loop' */
605 H = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
607 H = _mm_setzero_pd();
609 GMX_MM_TRANSPOSE2_PD(G,H);
610 Heps = _mm_mul_pd(gbeps,H);
611 Fp = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
612 VV = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
613 vgb = _mm_mul_pd(gbqqfactor,VV);
614 /* #define INNERFLOPS INNERFLOPS+10 */
616 /* #if 'Force' in KERNEL_VF */
617 FF = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
618 fgb = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
619 dvdatmp = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r{I}{J})));
620 /* #if ROUND == 'Epilogue' */
621 dvdatmp = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
623 dvdasum = _mm_add_pd(dvdasum,dvdatmp);
624 /* #if ROUND == 'Loop' */
625 gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
627 gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
629 /* #define INNERFLOPS INNERFLOPS+13 */
631 velec = _mm_mul_pd(qq{I}{J},rinv{I}{J});
632 /* #define INNERFLOPS INNERFLOPS+1 */
633 /* #if 'Force' in KERNEL_VF */
634 felec = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
635 /* #define INNERFLOPS INNERFLOPS+3 */
638 /* #elif KERNEL_ELEC=='Ewald' */
639 /* EWALD ELECTROSTATICS */
641 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
642 ewrt = _mm_mul_pd(r{I}{J},ewtabscale);
643 ewitab = _mm_cvttpd_epi32(ewrt);
644 eweps = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
645 /* #define INNERFLOPS INNERFLOPS+4 */
646 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
647 ewitab = _mm_slli_epi32(ewitab,2);
648 ewtabF = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
649 /* #if ROUND == 'Loop' */
650 ewtabD = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
652 ewtabD = _mm_setzero_pd();
654 GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
655 ewtabV = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
656 /* #if ROUND == 'Loop' */
657 ewtabFn = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
659 ewtabFn = _mm_setzero_pd();
661 GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
662 felec = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
663 /* #define INNERFLOPS INNERFLOPS+2 */
664 /* #if KERNEL_MOD_ELEC=='PotentialShift' */
665 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
666 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_sub_pd(rinv{I}{J},sh_ewald),velec));
667 /* #define INNERFLOPS INNERFLOPS+7 */
669 velec = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
670 velec = _mm_mul_pd(qq{I}{J},_mm_sub_pd(rinv{I}{J},velec));
671 /* #define INNERFLOPS INNERFLOPS+6 */
673 /* #if 'Force' in KERNEL_VF */
674 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
675 /* #define INNERFLOPS INNERFLOPS+3 */
677 /* #elif KERNEL_VF=='Force' */
678 /* #if ROUND == 'Loop' */
679 gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
682 gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
684 felec = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
685 felec = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
686 /* #define INNERFLOPS INNERFLOPS+7 */
689 /* #elif KERNEL_ELEC=='CubicSplineTable' */
691 /* CUBIC SPLINE TABLE ELECTROSTATICS */
692 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
693 /* #if ROUND == 'Loop' */
694 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
696 F = _mm_setzero_pd();
698 GMX_MM_TRANSPOSE2_PD(Y,F);
699 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
700 /* #if ROUND == 'Loop' */
701 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
703 H = _mm_setzero_pd();
705 GMX_MM_TRANSPOSE2_PD(G,H);
706 Heps = _mm_mul_pd(vfeps,H);
707 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
708 /* #define INNERFLOPS INNERFLOPS+4 */
709 /* #if 'Potential' in KERNEL_VF */
710 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
711 velec = _mm_mul_pd(qq{I}{J},VV);
712 /* #define INNERFLOPS INNERFLOPS+3 */
714 /* #if 'Force' in KERNEL_VF */
715 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
716 felec = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq{I}{J},FF),_mm_mul_pd(vftabscale,rinv{I}{J})));
717 /* #define INNERFLOPS INNERFLOPS+7 */
720 /* ## End of check for electrostatics interaction forms */
722 /* ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
724 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
726 /* #if KERNEL_VDW=='LennardJones' */
728 /* LENNARD-JONES DISPERSION/REPULSION */
730 rinvsix = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
731 /* #define INNERFLOPS INNERFLOPS+2 */
732 /* #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
733 vvdw6 = _mm_mul_pd(c6_{I}{J},rinvsix);
734 vvdw12 = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
735 /* #define INNERFLOPS INNERFLOPS+3 */
736 /* #if KERNEL_MOD_VDW=='PotentialShift' */
737 vvdw = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_{I}{J},_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
738 _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
739 /* #define INNERFLOPS INNERFLOPS+8 */
741 vvdw = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
742 /* #define INNERFLOPS INNERFLOPS+3 */
744 /* ## Check for force inside potential check, i.e. this means we already did the potential part */
745 /* #if 'Force' in KERNEL_VF */
746 fvdw = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
747 /* #define INNERFLOPS INNERFLOPS+2 */
749 /* #elif KERNEL_VF=='Force' */
750 /* ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
751 fvdw = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_pd(rinvsix,rinvsq{I}{J}));
752 /* #define INNERFLOPS INNERFLOPS+4 */
755 /* #elif KERNEL_VDW=='CubicSplineTable' */
757 /* CUBIC SPLINE TABLE DISPERSION */
758 /* #if 'Table' in KERNEL_ELEC */
759 vfitab = _mm_add_epi32(vfitab,ifour);
761 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
762 /* #if ROUND == 'Loop' */
763 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
765 F = _mm_setzero_pd();
767 GMX_MM_TRANSPOSE2_PD(Y,F);
768 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
769 /* #if ROUND == 'Loop' */
770 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
772 H = _mm_setzero_pd();
774 GMX_MM_TRANSPOSE2_PD(G,H);
775 Heps = _mm_mul_pd(vfeps,H);
776 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
777 /* #define INNERFLOPS INNERFLOPS+4 */
778 /* #if 'Potential' in KERNEL_VF */
779 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
780 vvdw6 = _mm_mul_pd(c6_{I}{J},VV);
781 /* #define INNERFLOPS INNERFLOPS+3 */
783 /* #if 'Force' in KERNEL_VF */
784 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
785 fvdw6 = _mm_mul_pd(c6_{I}{J},FF);
786 /* #define INNERFLOPS INNERFLOPS+4 */
789 /* CUBIC SPLINE TABLE REPULSION */
790 vfitab = _mm_add_epi32(vfitab,ifour);
791 Y = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
792 /* #if ROUND == 'Loop' */
793 F = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
795 F = _mm_setzero_pd();
797 GMX_MM_TRANSPOSE2_PD(Y,F);
798 G = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
799 /* #if ROUND == 'Loop' */
800 H = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
802 H = _mm_setzero_pd();
804 GMX_MM_TRANSPOSE2_PD(G,H);
805 Heps = _mm_mul_pd(vfeps,H);
806 Fp = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
807 /* #define INNERFLOPS INNERFLOPS+4 */
808 /* #if 'Potential' in KERNEL_VF */
809 VV = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
810 vvdw12 = _mm_mul_pd(c12_{I}{J},VV);
811 /* #define INNERFLOPS INNERFLOPS+3 */
813 /* #if 'Force' in KERNEL_VF */
814 FF = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
815 fvdw12 = _mm_mul_pd(c12_{I}{J},FF);
816 /* #define INNERFLOPS INNERFLOPS+5 */
818 /* #if 'Potential' in KERNEL_VF */
819 vvdw = _mm_add_pd(vvdw12,vvdw6);
820 /* #define INNERFLOPS INNERFLOPS+1 */
822 /* #if 'Force' in KERNEL_VF */
823 fvdw = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv{I}{J})));
824 /* #define INNERFLOPS INNERFLOPS+4 */
827 /* ## End of check for vdw interaction forms */
829 /* ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
831 /* #if 'switch' in INTERACTION_FLAGS[I][J] */
832 d = _mm_sub_pd(r{I}{J},rswitch);
833 d = _mm_max_pd(d,_mm_setzero_pd());
834 d2 = _mm_mul_pd(d,d);
835 sw = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
836 /* #define INNERFLOPS INNERFLOPS+10 */
838 /* #if 'Force' in KERNEL_VF */
839 dsw = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
840 /* #define INNERFLOPS INNERFLOPS+5 */
843 /* Evaluate switch function */
844 /* #if 'Force' in KERNEL_VF */
845 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
846 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
847 felec = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(velec,dsw)) );
848 /* #define INNERFLOPS INNERFLOPS+4 */
850 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
851 fvdw = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(vvdw,dsw)) );
852 /* #define INNERFLOPS INNERFLOPS+4 */
855 /* #if 'Potential' in KERNEL_VF */
856 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
857 velec = _mm_mul_pd(velec,sw);
858 /* #define INNERFLOPS INNERFLOPS+1 */
860 /* #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
861 vvdw = _mm_mul_pd(vvdw,sw);
862 /* #define INNERFLOPS INNERFLOPS+1 */
866 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
867 cutoff_mask = _mm_cmplt_pd(rsq{I}{J},rcutoff2);
868 /* #define INNERFLOPS INNERFLOPS+1 */
871 /* #if 'Potential' in KERNEL_VF */
872 /* Update potential sum for this i atom from the interaction with this j atom. */
873 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
874 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
875 velec = _mm_and_pd(velec,cutoff_mask);
876 /* #define INNERFLOPS INNERFLOPS+1 */
878 /* #if ROUND == 'Epilogue' */
879 velec = _mm_unpacklo_pd(velec,_mm_setzero_pd());
881 velecsum = _mm_add_pd(velecsum,velec);
882 /* #define INNERFLOPS INNERFLOPS+1 */
883 /* #if KERNEL_ELEC=='GeneralizedBorn' */
884 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
885 vgb = _mm_and_pd(vgb,cutoff_mask);
886 /* #define INNERFLOPS INNERFLOPS+1 */
888 /* #if ROUND == 'Epilogue' */
889 vgb = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
891 vgbsum = _mm_add_pd(vgbsum,vgb);
892 /* #define INNERFLOPS INNERFLOPS+1 */
895 /* #if 'vdw' in INTERACTION_FLAGS[I][J] */
896 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
897 vvdw = _mm_and_pd(vvdw,cutoff_mask);
898 /* #define INNERFLOPS INNERFLOPS+1 */
900 /* #if ROUND == 'Epilogue' */
901 vvdw = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
903 vvdwsum = _mm_add_pd(vvdwsum,vvdw);
904 /* #define INNERFLOPS INNERFLOPS+1 */
908 /* #if 'Force' in KERNEL_VF */
910 /* #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
911 fscal = _mm_add_pd(felec,fvdw);
912 /* #define INNERFLOPS INNERFLOPS+1 */
913 /* #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
915 /* #elif 'vdw' in INTERACTION_FLAGS[I][J] */
919 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
920 fscal = _mm_and_pd(fscal,cutoff_mask);
921 /* #define INNERFLOPS INNERFLOPS+1 */
924 /* #if ROUND == 'Epilogue' */
925 fscal = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
928 /* Calculate temporary vectorial force */
929 tx = _mm_mul_pd(fscal,dx{I}{J});
930 ty = _mm_mul_pd(fscal,dy{I}{J});
931 tz = _mm_mul_pd(fscal,dz{I}{J});
933 /* Update vectorial force */
934 fix{I} = _mm_add_pd(fix{I},tx);
935 fiy{I} = _mm_add_pd(fiy{I},ty);
936 fiz{I} = _mm_add_pd(fiz{I},tz);
937 /* #define INNERFLOPS INNERFLOPS+6 */
939 /* #if GEOMETRY_I == 'Particle' */
940 /* #if ROUND == 'Loop' */
941 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
943 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
945 /* #define INNERFLOPS INNERFLOPS+3 */
947 fjx{J} = _mm_add_pd(fjx{J},tx);
948 fjy{J} = _mm_add_pd(fjy{J},ty);
949 fjz{J} = _mm_add_pd(fjz{J},tz);
950 /* #define INNERFLOPS INNERFLOPS+3 */
955 /* #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
956 /* #if 0 ## This and next two lines is a hack to maintain indentation in template file */
961 /* ## End of check for the interaction being outside the cutoff */
964 /* ## End of loop over i-j interaction pairs */
966 /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
967 /* #if ROUND == 'Loop' */
968 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
970 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
972 /* #define INNERFLOPS INNERFLOPS+3 */
973 /* #elif GEOMETRY_J == 'Water3' */
974 /* #if ROUND == 'Loop' */
975 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
977 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
979 /* #define INNERFLOPS INNERFLOPS+9 */
980 /* #elif GEOMETRY_J == 'Water4' */
981 /* #if 0 in PARTICLES_J */
982 /* #if ROUND == 'Loop' */
983 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
985 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
987 /* #define INNERFLOPS INNERFLOPS+12 */
989 /* #if ROUND == 'Loop' */
990 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
992 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
994 /* #define INNERFLOPS INNERFLOPS+9 */
998 /* Inner loop uses {INNERFLOPS} flops */
1003 /* End of innermost loop */
1005 /* #if 'Force' in KERNEL_VF */
1006 /* #if GEOMETRY_I == 'Particle' */
1007 gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
1008 f+i_coord_offset,fshift+i_shift_offset);
1009 /* #define OUTERFLOPS OUTERFLOPS+6 */
1010 /* #elif GEOMETRY_I == 'Water3' */
1011 gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1012 f+i_coord_offset,fshift+i_shift_offset);
1013 /* #define OUTERFLOPS OUTERFLOPS+18 */
1014 /* #elif GEOMETRY_I == 'Water4' */
1015 /* #if 0 in PARTICLES_I */
1016 gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1017 f+i_coord_offset,fshift+i_shift_offset);
1018 /* #define OUTERFLOPS OUTERFLOPS+24 */
1020 gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1021 f+i_coord_offset+DIM,fshift+i_shift_offset);
1022 /* #define OUTERFLOPS OUTERFLOPS+18 */
1027 /* #if 'Potential' in KERNEL_VF */
1029 /* Update potential energies */
1030 /* #if KERNEL_ELEC != 'None' */
1031 gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1032 /* #define OUTERFLOPS OUTERFLOPS+1 */
1034 /* #if 'GeneralizedBorn' in KERNEL_ELEC */
1035 gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
1036 /* #define OUTERFLOPS OUTERFLOPS+1 */
1038 /* #if KERNEL_VDW != 'None' */
1039 gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1040 /* #define OUTERFLOPS OUTERFLOPS+1 */
1043 /* #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1044 dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai{I},isai{I}));
1045 gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
1048 /* Increment number of inner iterations */
1049 inneriter += j_index_end - j_index_start;
1051 /* Outer loop uses {OUTERFLOPS} flops */
1054 /* Increment number of outer iterations */
1057 /* Update outer/inner flops */
1058 /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1059 /* ## primitive and replaces aggressively even in strings inside these directives, we need to */
1060 /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source. */
1061 /* #if GEOMETRY_I == 'Water3' */
1062 /* #define ISUFFIX '_W3' */
1063 /* #elif GEOMETRY_I == 'Water4' */
1064 /* #define ISUFFIX '_W4' */
1066 /* #define ISUFFIX '' */
1068 /* #if GEOMETRY_J == 'Water3' */
1069 /* #define JSUFFIX 'W3' */
1070 /* #elif GEOMETRY_J == 'Water4' */
1071 /* #define JSUFFIX 'W4' */
1073 /* #define JSUFFIX '' */
1075 /* #if 'PotentialAndForce' in KERNEL_VF */
1076 /* #define VFSUFFIX '_VF' */
1077 /* #elif 'Potential' in KERNEL_VF */
1078 /* #define VFSUFFIX '_V' */
1080 /* #define VFSUFFIX '_F' */
1083 /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1084 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1085 /* #elif KERNEL_ELEC != 'None' */
1086 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1088 inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});